Abstract
The disclosure relates to a method for analyzing a plant, in particular for analyzing cannabis, using an illumination unit, a sensor unit, and an analysis unit, said analysis unit having a data-based classifier. The disclosure additionally relates to a device for analyzing a plant, said device comprising an illumination unit for lighting the plant to be analyzed and a sensor unit for receiving analysis input data, wherein the analysis input data contains at least spectral information, in particular an absorption spectrum or a reflection spectrum of the training plant. The device additionally comprises an analysis unit for analyzing the received analysis input data and for determining at least one property of the plant to be analyzed. The analysis unit is also designed to determine at least one property of the plant using a data-based classifier and the previously received analysis input data.
Claims
1. A method for analyzing a plant using an illumination unit, a sensor unit, and an evaluation unit, wherein the evaluation unit comprises a data-based classifier and the method comprises the following steps: training the classifier, the training comprising the following steps: illuminating a training plant having at least one known property using the illumination unit; acquiring training input data by measuring the radiation reflected from the training plant; training of the classifier with the acquired training input data, as well as training output data, wherein the training input data at least comprises spectral information including an absorption spectrum or a reflection spectrum of the training plant, and the training output data is associated to the training input data and comprise information about at least one property of the training plant; acquiring analysis input data by means of the sensor unit, wherein the acquiring comprises the following steps: illuminating a plant to be analyzed which has at least one unknown property, using the illumination unit; acquiring analysis input data by measuring the radiation reflected from the plant to be analyzed; determining a property of the plant to be analyzed, using the classifier previously trained with the trainings input data and the training output data and the acquired analysis input data.
2. The method according to claim 1, wherein for increasing the accuracy of the method, the training input data and the analysis input data additionally comprise images of the training plant and the plant to be analyzed.
3. The method according to claim 2, wherein the classifier additionally comprises the following steps: providing images as training input data which are to be used for training the classifier, and training output data associated to the images provided; rotating the provided images; associating the training output data, which are associated to the initially provided images, to the rotated images; combining the initially provided images and the rotated images, as well as the training output data associated to the images to an extended training data set; and training the classifier with the use of the extended training data set.
4. The method according to claim 1, wherein for increasing the accuracy of the method, the training input data and the analysis input data additionally include genetic information about plants.
5. The method according to claim 1, wherein the classifier is based on an artificial neural network, in particular a Convolutional Neural Network.
6. The method according to claim 1, wherein for increasing the accuracy of the method, during the training of the classifier, the training input data and the analysis input data include information about the temporal change in the input data acquired.
7. A device for analyzing a plant comprising: an illumination unit for illuminating the plant to be analyzed; a sensor unit for acquiring analysis input data, wherein the analysis input data include at least spectral information, including an absorption spectrum or a reflection spectrum of the plant to be analyzed; and an evaluation unit for evaluating the analysis input data acquired and for determining at least one property of the plant to be analyzed; wherein the evaluation unit is configured to determine the at least one property of the plant using a data-based classifier, as well as the previously acquired analysis input data.
8. The device according to claim 7, wherein the sensor unit comprises a spectrometer.
9. The device according to claim 7, wherein the sensor unit comprises a camera having a sensor surface.
10. The device according to claim 9, wherein the camera is designed as a 3D camera.
11. The device according to claim 7, wherein the illumination unit comprises at least two illumination elements.
12. The device according to claim 7, wherein the lighting elements are arranged on a circular path surrounding the sensor surface of the camera.
13. The device according to claim 7, wherein the sensor unit and/or the illumination unit comprise a cooling element.
14. The device according to claim 7, wherein the sensor unit additionally comprises a temperature sensor and/or a humidity sensor.
15. The device according to claim 7, wherein the sensor unit comprises a sensor surface provided with a UV/VIS conversion coating.
16. The method according to claim 1, wherein the plant is cannabis.
17. The device of claim 7, wherein the plant is a hemp plant.
Description
BRIEF DESCRIPTION OF THE DRAWINGS
[0094] The disclosure will be described in the following in more detail with reference to the embodiments and the drawings. Specifically, the drawings show the following
[0095] FIG. 1 is a schematic diagram of a first embodiment of the disclosure,
[0096] FIG. 2 is a schematic diagram of a second embodiment of the disclosure,
[0097] FIG. 3 is a schematic diagram of a third embodiment of the disclosure,
[0098] FIG. 4 is a top plan view on a first embodiment of the illumination unit of the analysis device of the present disclosure,
[0099] FIG. 5 is a top plan view on a second embodiment of the illumination unit of the analysis device of the present disclosure, wherein the sensor unit is integrated in the illumination unit,
[0100] FIG. 6 is a flow diagram illustrating the method of the present disclosure,
[0101] FIG. 7 is another flow diagram illustrating the individual steps of training the classifier,
[0102] FIG. 8 is a schematic diagram of a reflection spectrum acquired,
[0103] FIG. 9 is a schematic diagram of a first embodiment of the classifier, and
[0104] FIG. 10 is a schematic diagram of a second embodiment of the classifier.
DETAILED DESCRIPTION
[0105] FIG. 1 shows a first embodiment of the analysis device 10 of the present disclosure. The analysis device 10 comprises an illumination unit 12, a sensor unit 14, as well as an evaluation unit 16. The sensor unit 14 may, for example, be designed as a camera or a spectrometer. In the embodiment illustrated, the evaluation unit 16 is designed as a portable computer. The analysis device 10 is configured to analyze a plant (or a plant leaf) 18. To this end, the plant 18 is irradiated by the illumination unit 12. The radiation reflected from the plant 18 is subsequently captured by the sensor unit 14. The data acquired by the sensor unit 14 are evaluated by the evaluation unit 16 using a classifier which is not illustrated in this Figure. The sensor unit 14 is in particular configured to acquire spectral information. These may be acquired either directly, e.g., by a spectrometer, or also indirectly, e.g., by lighting the plant 18 with LEDs of different emission spectra and subsequently capturing the intensities using a RGB camera. This “scanning” of the spectrum may be performed in particular in the manner already described above. As such, it is not necessarily required that the sensor unit 14 comprises a spectrometer.
[0106] FIG. 2 illustrates a second embodiment of the analysis device 10 of the present disclosure. In this embodiment the sensor unit 14 is designed as a spectrometer 14a. The spectrometer 14a comprises a first lens L1, a diffraction grating G, a second lens L2, as well as a CMOS camera. The first lens L1 serves to collimate the light reflected from the plant 18. The collimated light then passes the diffraction grating G. The diffraction grating G decomposes the light into its spectral components and directs the components of different wavelengths to different regions of the CMOS camera. Thus, a spatial “spreading” of the light on the sensor surface of the CMOS camera is performed. by the subsequent evaluation of the image captured by the CMOS camera, it is therefore possible to determine the reflection spectrum of a plant 18. Optionally, in determining the reflection spectrum, information about the emission spectrum of the illumination unit 12 can be taken into consideration as well. The second lens L2 illustrated in FIG. 2 serves to again collimate the light spread by the diffraction grating G and diverging, so that the radiation arrives at the sensor surface of the CMOS camera.
[0107] FIG. 3 illustrates a third embodiment of the analysis device 10 of the present disclosure. In this embodiment, the sensor unit 14 comprises a spectrometer 14a for acquiring spectral information, a first 3D camera 14b for capturing the three-dimensional contour of the plant 18, a second 3D camera 14c serving to capture the three-dimensional contour of the plant 18 from an additional perspective, and an additional sensor 14d used to acquire environmental data. According to the embodiment illustrated in FIG. 3, the first 3D camera 14b can be arranged such that it captures the three-dimensional contour of the front of the plant 18 (or the plant leaf), while the second 3D camera 14c captures the three-dimensional contour of the rear of the plant (or the plant leaf) 18. In this manner, it becomes possible to determine the volume of the plant 18 and to calculate therefrom the absolute THC concentration, for example. The additional sensor 14d may, for example, measure the air temperature in the greenhouse or in the climate chamber and provide additional information for the classifier.
[0108] FIG. 4 is a top plan view on a first embodiment of the illumination unit 12. As can be seen in this Fig., the illumination unit 12 comprises a total of 12 LEDs 12a. The LEDs 12a may be identical LEDs, but may also be LEDs with different emission spectra. Here, narrowband-emitting LEDs can be used, but broadband LEDs could be used as well.
[0109] FIG. 5 illustrates a top plan view on a second embodiment of the illumination unit 12. In this embodiment, the sensor unit 14, which is designed as a camera and comprises a sensor surface 14e, is integrated in the illumination unit 12. The individual LEDs 12a are arranged on a circular path surrounding the sensor surface 14e. Thereby, it is possible to achieve a particularly homogeneous illumination of the plant 18 to be analyzed, as well as a particularly advantageous luminous efficiency of the camera.
[0110] FIG. 6 is an overview of the method of the present disclosure. First, the classifier, which is based in particular on a neural network, is trained using a training data set (step S100). This training data set includes spectral information to which specific properties of a plant 18 can be associated. For example, the classifier may learn during the training phase that an absorption peak at 350 nm indicates that the plant 18 to be analyzed is infested by a certain disease. The classifier could also learn during the training phase that measuring two absorption peaks, of which the first peak is detected at 320 nm and the second peak is detected at 480 nm, indicates that the plant is healthy and has a high water content. The classifier can also be trained to take further input data into account, in particular additional images or information about temperature, humidity, light conditions, genetic information about the plant and/or the plant varieties. The more training data is available during the training phase, the more reliable the classifier can evaluate the plant to be analyzed at a later time. After the training of the classifier (step S100), the method of the present disclosure is ready for implementation.
[0111] For the analysis of a plant 18, first, the analysis input data are acquired (step S200). The acquisition of the analysis input data is performed essentially in the same manner as the acquisition of the training input data, which will be discussed in the context of FIG. 7. As soon as the analysis input data, in particular an absorption spectrum, have been acquired, the plant property is determined (step S300) using the previously trained classifier.
[0112] FIG. 7 illustrates the individual sub-steps of the training of the classifier (step S100). First, a training plant having known properties is illuminated using the illumination unit 12 (step S110). Subsequently, the training data, specifically the training input data, are acquired by means of the sensor unit 14 (step S120). The training input data are then associated to the predetermined and already known training output data which include information about the property of a plant. In this manner, the classifier “learns” which output data or properties correlate with which input data. After the correlator has been trained (step S130), the method and the classifier are ready for implementation.
[0113] FIG. 8 is an exemplary illustration of a reflection spectrum acquired. This reflection spectrum may have been acquired during the analysis process, for example. The reflection spectrum is the result of a spectral analysis of the light reflected from a plant. In the reflection spectrum illustrated, three different points are identified as P1, P2 and P3. These points are local minima of the reflection spectrum and local maxima of a corresponding absorption spectrum. These local extreme values may be searched for and determined in predeterminable regions, for example. It may be provided, for example, that the extreme values are determined in the wavelength ranges b1=200 to 400 nm, b2=400 to 600 nm and b3=600 to 800 nm. The result of a corresponding analysis may be, for example, that a first absorption peak is determined at P1=280 nm, a second absorption peak is determined at P2=550 nm, and a third absorption peak is determined at P3=790 nm. These three values can thus be used as analysis input data to determine that the plant analyzes is of the variety Sativa and is also healthy.
[0114] The basic functionality of the classifiers used in the context of the disclosure is illustrated in FIGS. 9 and 10 in an abstracted manner.
[0115] FIG. 9 illustrates the case that the three features P1, P2 and P3 are determined during the analysis of a plant. As describes above, these features may be three absorption peaks, for example. In the example illustrated in FIG. 9, the features P1 to P3 are used as input data for the classifier to conclude on an output data item Q. For example, Q may relate to the state of the plant (e.g., diseased or healthy) or to the THCC concentration.
[0116] Finally, FIG. 10 illustrates the case that the three features P1, P2 and P3 are used as input data to analyze a plant, and the quantities Q1, Q2 and Q3 represent the output data. As explained in the above examples, P1 to P3 may describe three local absorption peaks. The three output quantities Q1 to Q3 may describe three properties of a plant. For example, Q1 may describe the plant variety, whereas Q2 describes the health state of a plant and Q3 describes the water content of the plant, for example.
